MDS Parameters 

Home
About Clustan
Cluster Analysis
Applications
ClustanGraphics
User Support
Clustan/PC
Orders
What's New
White Papers
Contact Us
MDS Parameters
Main dialogue for ClustanMDSClustanMDS needs only a proximity matrix as input.  It can be either similarities or dissimilarities, and can be read in using File/New/Proximities or computed using Prox/Compute.

To run ClustanMDS, just click MDS on the Prox menu.  The dialogue shown left will appear, with the size of the proximity matrix indicated in the first line and the other parameter values suggested as defaults.  If you are happy with the defaults, click start and ClustanMDS will run.

Note that ClustanMDS took less than half a second to complete 100 MDS trials on the Mammals milk data, selecting the best solution from trial 64 with stress < 1%, which Kruskal rates as an excellent fit.  The stress value means that over 99% of the distances between the points on the scatterplot match the rank order of the dissimilarities in the proximity matrix.

Click "Details" to obtain information on the results obtained in the 100 trials, or click "Finish" and your final configuration will be plotted - full example here.

The parameters for ClustanMDS are as follows:

Number of trials >> the number of times that ClustanMDS is to be run, each trial starting with a different random initial configuration.

Number of iterations >> the maximum number of hill-climbing iterations to be performed in each trial.  ClustanMDS will often stop before this number is reached, for example if there is no improvement in the stress value in successive iterations.  However, a limit must be specified if the procedure fails to converge.

Number of dimensions >> is the number of new variables to be saved by ClustanMDS.  The standard default is 2, so that you obtain the best two-dimensional representation of your proximity matrix.  However, it can be changed to 1, which gives the best linear fit; or a number higher than 2, if you think your data is more complex than two-dimensional.

The output configuration from ClustanMDS is saved as new scatterplot variables, which can be graphed using Cluster/Scatterplots or displayed and copied using View/Data.